Tax Professionals’ Perceptions, Compliance Costs, and Compliance Intentions Under Indonesia’s Core Tax Administration System

Abstract

This study provides an early evaluation of the effectiveness of the Core Tax Administration System, a digital taxation platform introduced to integrate all tax administration processes in Indonesia into a single system. To conduct this evaluation, the study integrates two of the most established frameworks in the information systems literature, namely the DeLone and McLean Information Systems Success Model and the Technology Acceptance Model. Tax professionals are involved in the evaluation process because they are the primary users of the system and possess advanced knowledge of taxation. Structural equation modeling is employed as the analytical technique. The results indicate that system usage generates individual-level benefits by reducing perceived compliance costs, which in turn translate into organizational-level outcomes in the form of increased tax compliance intentions. However, the non-linear effect analysis reveals that this relationship is not entirely linear but follows an inverted U-shaped pattern. This finding suggests that over time, highly routine system usage may reduce professional vigilance by fostering excessive reliance on automated features and superficial processing. Such dependence can weaken perceived efficiency gains and diminish intrinsic motivation for careful and accurate reporting, highlighting the importance of balancing efficiency with system design features that support professional judgment and vigilance.

1. Introduction

Over the past decade, Indonesia’s tax-to-gross domestic product (GDP) ratio has averaged below 11%, considerably lower than the Asia-Pacific regional average of around 19% [1]. This relatively low ratio may reflect a combination of factors, including a relatively narrow tax base, policy design features, and challenges related to tax compliance. The World Bank [2] suggests that a tax ratio threshold of 15% represents a critical tipping point, beyond which tax revenues become an essential prerequisite for sustainable economic growth. This perspective is further supported by Gaspar et al. [3], who demonstrate that countries surpassing this 15% benchmark achieve per capita income levels 7.5% higher than projected within a ten-year horizon. In essence, adequate tax revenues provide governments with the fiscal space to finance development investments without excessive reliance on debt, thereby fostering long-term growth.

Acknowledging the urgency of this challenge, President Prabowo Subianto—sworn into office in October 2024—set a target of raising the tax-to-GDP ratio to between 11.52% and 15% by 2029, as stipulated in Presidential Regulation No. 12 of 2025 on the National Medium-Term Development Plan 2025–2029 [4]. Achieving this target, however, remains highly demanding given the substantial gap from current performance (around 10.03%). It requires policy interventions that strengthen compliance by addressing inefficiencies in tax administration, thereby enabling revenues to expand independently of GDP growth. Without such reforms, simultaneous increases in GDP and tax revenues may merely offset each other, leading the tax ratio to stagnate or even decline [5].

This concern is supported by empirical evidence from Tandjung [6], who found that a 1% increase in nominal GDP in Indonesia generated only a 0.93% to 1.03% increase in tax revenues. More recent evidence by Sinaga et al. [7], using quarterly data from 2015 to 2021 across 34 provinces, reports an aggregate tax buoyancy of 1.25. At the disaggregated level, personal income tax exhibited the highest buoyancy (1.35), followed by value-added tax (1.28), while corporate income tax lagged with a coefficient of 0.937. These findings indicate that Indonesia has never achieved a condition in which tax revenues grew at twice the pace of nominal GDP growth.

Against this backdrop, strengthening technology-driven tax administration has become a strategic national priority [5]. A milestone in this agenda was the official launch of the Core Tax Administration System (CTAS) by the Directorate General of Taxes (called ‘DJP’ in Indonesia) on 1 January 2025 (see https://coretaxdjp.pajak.go.id/ (accessed on 21 January 2026)). As the technological backbone of Tax Reform Volume III [8], CTAS is mandated under Ministry of Finance Regulation No. 81 of 2024 and is designed to integrate the entire core processes of tax administration, including taxpayer registration, tax return (called ‘SPT’ in Indonesia) filing, payment, audit, and collection. Compared with its predecessor, e-Filing and e-Invoice platform (see https://www.pajak.go.id/en/node/41026 (21 January 2026)), CTAS introduces a more advanced and fully integrated architecture, expected to deliver faster, more transparent, and more accurate services that ultimately reduce compliance costs and strengthen compliance [9]. Optimism regarding its potential has also been highlighted in academic discourse, which positions CTAS as a key driver for raising Indonesia’s tax ratio to a more sustainable level [10].

Nonetheless, classical challenges commonly associated with newly implemented systems surfaced soon after the CTAS rollout. Numerous users, particularly taxpayers, reported severe technical difficulties, most notably server capacity constraints that limited access to the official portal and obstructed timely compliance [11]. The issue quickly escalated into a broader public debate as media coverage highlighted potential financial losses from administrative delays, particularly in relation to sanctions [12]. In response, the DJP introduced transitional relief measures exempting taxpayers from administrative penalties for late filing and payment, with the transition period set at three months after the system’s launch, as stipulated in Ministry of Finance Regulation No. 131 of 2024.

However, instead of demonstrating a firm commitment to improving CTAS, the government appeared to retreat by implicitly acknowledging the system’s immaturity. This was reflected in the DJP’s decision to reactivate the e-Invoice platform—previously limited to taxpayers issuing at least 10,000 invoices per month under Decree No. KEP-24/PJ/2025—for use by all taxable entrepreneurs. The policy, formalized in DJP Decree No. KEP-54/PJ/2025 and effective 12 February 2025, was intended to facilitate invoice preparation during the transition period. The situation has been further complicated by the absence of a clear timeline for ending this arrangement, as persistent lobbying from the business community has pushed for both an extension of the relief period and the elimination of penalties in light of unresolved CTAS challenges. Consequently, policy execution has become inherently contradictory: while the government urges taxpayers to embrace a digital tax era, the inconsistent implementation of CTAS perpetuates uncertainty and undermines public confidence in the reform. As Merton [13] famously argued through the concept of the self-fulfilling prophecy, the absence of public confidence can equally set in motion a reality of failure.

The negative issues surrounding the CTAS implementation, as discussed above, have raised public scepticism about the system’s ability to achieve its objectives, particularly in reducing compliance costs and fostering taxpayer compliance through the modernization of tax administration [14]. This context underscores the importance of conducting an early evaluation of CTAS effectiveness, especially considering the substantial investment of approximately more than IDR 1 trillion required for its development [15]. Such large-scale expenditures naturally create high expectations among taxpayers regarding the success of CTAS implementation. As a specific form of information system (IS), evaluating CTAS performance entails identifying the factors that influence primary users (taxpayers) in adopting and continuing to use the system [16]. This perspective is reflected in the IS success model developed by DeLone and McLean [17], which highlights constructs such as system quality, information quality, and user satisfaction as key determinants of effective IS adoption. Through these mechanisms, the expected benefits of IS implementation—such as reduced compliance costs and improved tax compliance—can be realized.

Furthermore, as highlighted by Saragih et al. [18], CTAS represents a significant advancement in tax technology. In this regard, the Technology Acceptance Model (TAM), developed by Davis [19], provides a relevant conceptual framework for assessing taxpayers’ responses to such technological innovations. Similar to the DeLone and McLean (D&M) model, TAM emphasizes that the success of technology implementation can only be evaluated through user experience. According to TAM, an individual’s intention to adopt a technology is primarily influenced by their perceptions of ease of use and perceived usefulness. Consequently, if taxpayers encounter difficulties in operating CTAS, it is likely to hinder actual system utilization. As a result, the substantial investment allocated to the development of this system risks becoming ineffective.

Therefore, this study aims to quantitatively analyze taxpayers’ perceptions of the success of CTAS implementation, based on their actual experience in using the system. By integrating the D&M model with the TAM, the research focuses on identifying the factors that influence taxpayers’ decisions to continue utilizing the system in fulfilling their tax obligations. In addition, the study explores the extent to which CTAS can enhance taxpayers’ propensity to comply with tax regulations. The taxpayers under consideration are tax professionals, including tax consultants and tax accountants, who not only fulfil their own tax obligations but also assist clients or the companies they represent in meeting theirs. This group constitutes the primary users of CTAS and possesses deeper insights into tax administration services compared to the general taxpayer population. Consequently, their perspectives provide a valid basis for assessing the issues under investigation and play a critical role in informing the design of future tax policy in Indonesia [20].

The novelty of this study lies in the integration of the D&M model and TAM into an innovative framework that comprehensively evaluates the effectiveness of tax administration system implementation. Both models have been extensively applied across various domains of IS, such as management [21], education [22], public administration [23], and marketing [24]. However, the integration of TAM and the D&M model within the taxation context remains underexplored. Addressing this gap is therefore essential, as it has the potential to provide valuable insights for tax authorities in developing more effective tax administration systems that better meet taxpayers’ needs. In this regard, the introduction of the CTAS initiative—aimed at streamlining tax compliance processes—represents a noteworthy reform effort. Nevertheless, from the taxpayers’ perspective, a positive attitude toward the tax authority can only emerge if the administrative system is perceived as responsive and user-friendly [25]. Hence, evaluating the effectiveness of the transition toward a more modern and integrated digital tax system, such as CTAS, becomes critically important.

A related study by Saptono et al. [26] examined Indonesia’s e-Filing and e-Form services but focused solely on the perceptions of tax consultants. In contrast, the present study extends the analysis to a broader user group—namely tax accountants—who often report challenges in operating CTAS. Understanding their perceptions and satisfaction levels is therefore pivotal for fostering a more reliable and compliance-oriented tax environment.

Accordingly, this study seeks to answer the following research questions: How do system quality, information quality, and service quality influence user satisfaction, system usage, and perceived compliance costs in the context of Indonesia’s CTAS, and how do these factors relate to tax compliance intentions among tax professionals?

The remainder of this paper is organized as follows. Section 2 provides a review of the literature, focusing on the digital transformation of tax administration in Indonesia and the theoretical foundations applicable for evaluating its effectiveness. Section 3 introduces the conceptual model and formulates the hypotheses that guide the empirical investigation. Section 4 outlines the research methodology, including data sources, questionnaire development, and analytical techniques employed to generate empirical evidence. Section 5 reports the empirical results along with robustness assessments addressing potential issues of nonlinearity, endogeneity, and unobserved heterogeneity. Section 6 presents an in-depth discussion of the findings, highlighting both theoretical and practical implications. Finally, Section 7 concludes the paper by summarizing the key contributions and suggesting directions for future research.

2. Background

2.1. The Digital Transformation of Tax Administration in Indonesia

The rapid advancement of information and communication technology has heightened societal demands for bureaucratic systems that provide services that are clear, timely, transparent, and accessible in real time [27]. Consequently, the conventional view of bureaucracy—often characterized by complexity and protracted procedures—must be urgently reformed to meet these emerging expectations [28]. Such transformation is expected to strengthen governance and generate multiple positive outcomes, including enhanced public service delivery, increased trust in governmental performance, improved regulatory compliance, stimulated investment, and accelerated economic growth [29,30].

Indonesia has demonstrated a strong commitment to bureaucratic transformation, as evidenced by the establishment of the National Ombudsman Commission through Presidential Decree No. 44 of 2000. This institution was created to enhance oversight of the state apparatus and rebuild public trust in government following the New Order era [31]. Building on this foundation, the government has prioritized the digitalization of the bureaucracy to create a clean, transparent, accountable, and corruption-free administration [32,33,34]. On 9 June 2003, the government issued Presidential Instruction No. 3 of 2003, which provides the legal framework for the implementation of e-government initiatives [35]. One of the earliest applications of e-government in Indonesia was e-procurement, introduced on 20 April 2005 under the Regulation of the Minister of Public Works No. 207 of 2005, which facilitated the electronic procurement of public goods and services [36]. The ongoing bureaucratic transformation remains a key focus in Indonesia and is included as one of the primary agenda items in the National Long-Term Development Plan 2025–2045 [37].

The digital transformation of tax administration, aimed at simplifying processes for taxpayers to fulfil their obligations, constitutes a vital element of bureaucratic reform in Indonesia [38]. This country is known for having a particularly complex tax system, with the tax complexity index in 2022 recorded at 0.44, ranking 56th out of 64 countries [39]. This score is higher than several Southeast Asian peers, such as Malaysia (0.39) and Singapore (0.26). According to Awasthi and Bayraktar [40], complex tax systems often result in arbitrary tax collection practices, which in turn create opportunities for corruption and foster tax evasion. Consequently, Indonesia has persistently experienced relatively low and stagnant tax compliance [41], as reflected in its tax-to-GDP ratio of only 10.3% in 2023—down from 10.4% in the previous year—compared to 12.6% in Malaysia and 13.9% in Singapore [42]. This configuration of challenges underscores the urgent need for a more extensive digital transformation in Indonesia’s tax administration to reduce compliance burdens and enhance tax compliance amid an increasingly intricate tax system [43].

Law No. 1 of 2004 on State Treasury represents the foundation for the digitalization of tax administration in Indonesia, as it grants the Minister of Finance the authority to establish procedures for state financial management, including the development of an electronic-based revenue system. A key strategic initiative identified as a ‘quick win’ is the launch of an e-Filing, which facilitates online submission of annual income tax returns for both individuals and corporations. This platform was officially launched on 14 May 2004 through DJP Decree No. KEP-88/PJ/2004, and its operational procedures were subsequently detailed in DJP Regulation No. KEP-05/PJ/2005. Under this regulation, taxpayers could submit their annual SPT electronically through application service providers designated by the DJP. Following the implementation of the State Revenue Module in early 2007, DJP developed its own e-Filing application, which was made available in 2014 through DJP Online. This platform integrated both e-Filing and payment services (e-Billing).

From the perspective of value-added tax (VAT), the digitalization of tax administration in Indonesia has been advanced through the introduction of the electronic tax invoice, or e-Invoice [44]. This platform enables the electronic issuance of VAT invoices and the online filing of monthly VAT returns, thereby ensuring faster and more transparent reporting. Its development was partly motivated by the inefficiencies of paper-based VAT administration, which required taxpayers to maintain detailed transaction records and submit periodic reports to the tax authority [45]. Such complexity encouraged some taxpayers to circumvent the system, for instance by issuing fictitious tax invoices to unlawfully claim input VAT credits, thereby posing risks to state revenue [46,47].

The implementation of e-Invoice was carried out in a step-by-step fashion, as stipulated under DJP Regulation No. PER-16/PJ/2014 and DJP Decree No. KEP-136/PJ/2014. The initial step began on 1 July 2014 with 45 selected taxable entrepreneurs, followed by an expansion to all taxable entrepreneurs within DJP’s Large Taxpayer Regional Offices as well as those in Java and Bali on 1 July 2015, before its nationwide adoption on 1 July 2016. Since then, e-Invoice has undergone several updates, culminating in the launch of e-Invoice 4.0 on 20 July 2024. The latest version introduced a prepopulated feature that allows VAT returns to be automatically generated from DJP’s system, thereby reducing the need for manual entry, minimizing potential data-input errors, and accelerating the reporting process.

Subsequently, the DJP undertook an integrative reform with the launch of the CTAS on 1 January 2025. CTAS consolidates key functions of tax administration that were previously managed through separate platforms such as e-Filing, e-Billing, e-Invoice, and other digital services. The system also encompasses a broad spectrum of DJP’s operational activities—from taxpayer registration to data analytics—thereby enabling more effective monitoring of taxpayer compliance. With this integrated platform, taxpayers are no longer required to navigate multiple systems to access administrative services. CTAS is thus anticipated to provide a long-term solution for simplifying, accelerating, and automating tax compliance processes. This initiative is particularly crucial in reinforcing tax compliance amid increasingly complex and digitalized global economic dynamics. Public expectations for its positive impact have intensified, especially following reports that the development cost of CTAS reached IDR 1.3 trillion—exceeding even the investment allocated to the development of advanced artificial intelligence technologies such as ChatGPT (IDR 1 trillion) and DeepSeek (IDR 97 billion) [48].

Since the CTAS only commenced recording tax transactions in 2025, the filing of annual income tax returns for 2024—given that these transactions were not captured in the system—was still conducted through the conventional channel, namely DJP online. It implies that, throughout 2025, CTAS utilization has been limited primarily to withholding and collecting agents, typically consisting of business taxpayers and employers, rather than individual taxpayers at large. Accordingly, the main functions available in CTAS to date include the electronic issuance of VAT invoices, the generation of withholding tax certificates, and the filing of monthly VAT and income tax returns. According to the DJP [49], as of April 2025, CTAS had already administered approximately 198.85 million VAT invoices, 70,693,689 withholding tax certificates, 933,484 monthly VAT returns, and 997,705 monthly income tax returns. Based on these figures, the DJP reported that the performance of the CTAS application system has demonstrated overall stability, although fluctuations in latency persist, particularly during periods of significantly increased transaction volumes in certain functions.

In practice, however, the implementation of CTAS has been met with considerable challenges. Numerous complaints have surfaced on social media platforms [50], particularly from tax accountants—individuals responsible for assisting corporations in fulfilling their tax obligations [51]. Faced with persistent technical and operational difficulties, many taxpayers have resorted to outsourcing compliance tasks to professional tax consultants [52]. This shift has, in turn, imposed additional compliance costs in the form of professional service fees, raising concerns about the unintended economic burden generated by the system’s initial rollout [53]. These development issues underscore the importance of conducting a more rigorous empirical evaluation of CTAS’s early implementation, not only to identify the extent of these challenges but also to capture the perceptions and experiences of its primary users. Such an evaluation is critical for assessing whether CTAS, which has been promoted as a landmark in Indonesia’s tax administration reform, is effectively meeting its intended objectives or instead creating new layers of complexity for taxpayers.

2.2. Theoretical Foundation

Evaluating the performance of CTAS requires a solid theoretical foundation. The TAM and the D&M model provide relevant frameworks for this purpose. CTAS represents a specialized form of IS innovation, making user perceptions and experiences central to its evaluation. As noted by Strong et al. [16], the success of an IS cannot be assessed independently from the experiences of its users.

The TAM, introduced by Davis [19], explains how users adopt new technologies. The model proposes that perceived usefulness and perceived ease of use are the primary determinants of technology acceptance. These perceptions shape behavioral intention, which subsequently influences actual system use [54,55,56]. Due to its parsimony and explanatory power, TAM has been widely used to examine technology adoption across various contexts [57]. Figure 1 illustrates the model.

Despite its widespread application, TAM has been criticized for its narrow focus, as it evaluates only a limited aspect of IS performance, overlooking broader dimensions such as the nature of use, service quality, and individual or organizational impacts [58]. To address these limitations, DeLone and McLean [59] introduced a more comprehensive model for assessing IS effectiveness (D&M model).

Initially, the D&M model identified six key factors: system quality, information quality, user satisfaction, utilization, individual impact, and organizational impact (see Figure 2). However, it faced criticism and underwent revisions within its first decade. Notably, Pitt et al. [60] highlighted the need to include service quality to enhance the evaluation of IS performance and minimize measurement errors. This perspective was further supported by Kettinger and Lee [61] and Li [62].

Responding to these critiques, DeLone and McLean [17] revised their model a decade later (see Figure 3). The revised framework incorporated service quality as a critical determinant of IS success and introduced refinements such as mediating variables to capture the relationship between user satisfaction and system use, thereby accommodating both obligatory and discretionary usage contexts. Furthermore, all impact measures—whether individual or organizational—were consolidated into the category of net benefits, underscoring the broader advantages of IS implementation. This reclassification highlights that the value of IS extends beyond immediate users, potentially influencing macro-level outcomes such as national economic performance.

Numerous studies have employed the TAM and D&M models to comprehensively evaluate IS performance across diverse contexts. For example, Chung et al. [21] used survey data from the United States to develop a model assessing the effectiveness of construction industry-specific Enterprise Resource Planning (ERP) systems. By integrating elements of TAM and the D&M framework, their model demonstrated that ERP adoption is shaped not only by technical attributes such as system functionality and reliability but also by organizational factors, including user acceptance and process integration.

In a different context, Mohammadi [22] examined e-learning adoption in Iran through the combined application of TAM and the D&M model. They revealed that both user intention and satisfaction significantly influenced effective e-learning utilization, with system quality and information quality emerging as key antecedents, and perceived usefulness mediating the relationship between ease of use and intention. Similarly, Napitupulu [23] designed an integrated e-government adoption framework for Indonesia by drawing on TAM, the D&M model, and complementary theories. In the financial sector, Mansour [24] investigated the determinants of consumer willingness to adopt Sharia-compliant mobile banking in Palestine, showing that the integrated application of TAM and the D&M model accounted for substantial variance in user adoption intentions.

Although applied in different domains, the underlying rationale for integrating the two models remains consistent. TAM captures individual adoption behavior, while the D&M model extends the evaluation to both individual and organizational impacts.

However, the application of these frameworks in the taxation domain presents distinctive theoretical considerations. Unlike many IS examined in prior TAM and D&M studies, digital tax administration systems operate within a regulatory environment in which system interaction is closely linked to legally mandated compliance obligations. Consequently, users engage with such systems not purely for efficiency or productivity purposes, but also as part of fulfilling statutory responsibilities.

This institutional setting introduces an additional dimension rarely incorporated into conventional IS adoption, namely the role of compliance-related outcomes, such as perceived compliance costs. By integrating TAM and the D&M model within this regulatory context, the present study extends the application of these frameworks beyond voluntary technology adoption settings and situates them within a compliance-oriented institutional environment. In doing so, the proposed framework links IS success factors with behavioral intentions related to tax compliance, thereby offering a theoretically enriched perspective on the role of digital tax administration systems.

Empirical evidence demonstrates that computerized taxation not only enhances compliance but also improves corporate financial performance [63], making this effort particularly relevant for CTAS, which is currently limited to taxable entities. Moreover, by integrating TAM and the D&M model, this study can provide valuable insights for tax authorities in identifying areas for improvement. In turn, such findings could contribute to the development of a more user-friendly and efficient tax system, ultimately benefiting both tax authorities and taxpayers.

Building on the theoretical perspectives discussed above, the present study develops a conceptual framework that integrates TAM and the D&M model to evaluate the performance of CTAS and its implications for tax compliance. Although more recent models, such as the Unified Theory of Acceptance and Use of Technology (UTAUT) and UTAUT2 extend the original TAM framework by incorporating additional social and contextual factors, this study adopts TAM due to its parsimony and its conceptual compatibility with the D&M model. Given that the present research already integrates multiple IS success constructs—such as system quality, information quality, service quality, satisfaction, and system use—using TAM provides a theoretically grounded yet parsimonious mechanism for capturing core technology acceptance perceptions without introducing excessive model complexity. The development of the conceptual model and the corresponding hypotheses is presented in the following section.

3. Conceptual Model and Hypothesis Development

Based on the theoretical foundations discussed above, several key points merit attention. First, within TAM, perceived ease of use and perceived usefulness of an IS are identified as independent variables. Similarly, in the D&M model, the three dimensions used to measure IS quality—system quality, service quality, and information quality—are also considered independent variables.

Second, this set of independent variables plays a critical role in shaping users’ willingness to adopt, which in the D&M model is further amplified by the level of user satisfaction. Third, whether an IS is actually adopted largely depends on usage intention, as emphasized in both TAM and the D&M model. Lastly, once adopted, an IS is expected to generate benefits for both individuals and organizations.

In the context of CTAS, individual-level benefits are reflected in the reduction in perceived compliance burdens among tax accountants (individuals authorized to manage corporate tax obligations) and tax consultants (professionals assisting clients such as taxable entrepreneurs and corporate entities in fulfilling their responsibilities). A reduction in these burdens, in turn, is expected to translate into higher levels of corporate tax compliance within the organizations that tax accountants serve and in the advisory work performed by tax consultants. This constitutes the organizational-level benefit of CTAS. Such a logical pathway is particularly relevant given that, during the study period (2025), CTAS was accessible exclusively to taxable entrepreneurs and corporate taxpayers. Accordingly, our conceptual model is grounded in these propositions and is presented in Figure 4.

Based on the constructed model, the definitions of each variable are formulated with reference to the existing theoretical literature. Establishing these definitions is essential for developing a shared understanding of the terms and constructs under investigation, thereby ensuring the consistency and validity of the research [64]. Furthermore, by clearly delineating these definitions, the study can frame its hypotheses with greater precision and situate its findings in relation to prior research. In doing so, this not only enhances the credibility of the research but also facilitates the replication of studies and contributes to the cumulative knowledge in the field [65]. Concise definitions of the variables and the hypotheses regarding their interrelationships within the model are presented below.

3.1. System Quality

System quality within an IS is commonly assessed through three key indicators, namely user-friendliness, ease of use, and performance reliability [58,59,66]. These indicators capture the technical success of the system in enabling taxpayers to fulfil their obligations with minimal obstacles [17]. In this study, system quality refers to the capacity of the CTAS to meet taxpayer requirements by ensuring seamless navigation across its multiple modules, including registration, filing, payment, and account management. Key attributes include accessibility, clear guidance, stable system performance, user-oriented design, and flexibility in input and modification of data. When taxpayers perceive the system as reliable and useful, it reduces compliance burdens, improves efficiency in fulfilling tax obligations, and ultimately enhances satisfaction in using CTAS [67]. Accordingly, the following hypothesis is proposed:

H1. 

System quality has a positive effect on user satisfaction.

3.2. Service Quality

According to Zeithaml et al. [68], DeLone and McLean [17], and DeLone and McLean [58], the assessment of IS service quality is based on four key indicators: readiness for service, safe transactions, availability, and empathy. DeLone and McLean [17] further clarify that empathy refers to prioritizing the best interests of users by providing individual attention and accommodating their specific needs. Pitt et al. [60] argue that measurement errors in evaluating IS effectiveness can be avoided by considering not only product-focused variables but also customer relationship–oriented factors, such as service quality. In this study, service quality relates to the extent of assistance provided by tax officers to facilitate taxpayers’ use of the CTAS. In the Indonesian context, the tax authority offers various support mechanisms, such as the Kring Pajak or call center, to help taxpayers comply with their filing obligations. Empirical evidence also supports the importance of this construct. For instance, Goh et al. [69] and Rahman et al. [70] found that improvements in service quality are positively associated with higher levels of user satisfaction with IS. Therefore, the following hypothesis is formulated:

H2. 

Service quality has a positive effect on user satisfaction.

3.3. Information Quality

Information quality within an IS encompasses several critical attributes, including accuracy, timeliness, completeness, reliability, and value or meaningfulness [59,71]. In contrast to system quality, which primarily reflects technical success, information quality captures the semantic success of an IS [17]. As emphasized by Mingers [72], achieving success at this level provides the most substantive foundation for the further development and long-term effectiveness of an IS. In the context of this study, information quality refers to the effectiveness of information provided by the CTAS in assisting taxpayers in meeting their tax obligations. Enhancing the system’s capacity to deliver high-quality information is expected to strengthen users’ perceptions of its informational value. Prior studies by Colesca and Dobrica [73] and Wang and Liao [74] empirically demonstrate that the availability of high-quality information improves taxpayers’ efficacy in utilizing IS, thereby fostering greater user satisfaction. Accordingly, the following hypothesis is proposed:

H3. 

Information quality has a positive effect on user satisfaction.

3.4. Perceived Ease of Use

Perceived ease of use refers to the extent to which an individual believes that using a particular system requires minimal effort [19]. It implies that users find it easy to interact with the system, which in turn facilitates their ability to recall how tasks are performed and enhances their skills in system usage [75]. This construct is conceptually comparable to system quality in the D&M model; however, it does not fully capture the technical success of an IS [17]. Despite this distinction, perceived ease of use has consistently been identified as a critical determinant of individuals’ willingness to adopt new technology-based applications [76]. In the taxation context, prior studies highlight similar findings. For instance, taxpayers’ perceptions of the ease of using e-Filing systems have been shown to significantly influence their intention to adopt such platforms [77,78,79]. Moreover, perceived ease of use is believed to exert an indirect influence on usage intentions by enhancing perceived usefulness of the system [80,81]. Thus, perceived ease of use not only reduces barriers to system adoption but also strengthens the perceived value of the system in improving task performance. Accordingly, the following hypotheses are proposed:

H4. 

Perceived ease of use has a positive effect on intention to use.

H5. 

Perceived ease of use has a positive effect on perceived usefulness.

3.5. Perceived Usefulness

Perceived usefulness refers to the degree to which an individual believes that using a specific system will enhance their job performance [19]. In essence, individuals are more likely to adopt or reject an application depending on the extent to which they perceive it as beneficial in accomplishing their tasks more effectively. This perception is typically evaluated through improvements in task effectiveness, work quality, and overall productivity that result from system use [75]. These measures are conceptually comparable to the information quality dimension of the D&M model and overlap with several factors in the user information satisfaction models developed by Ives et al. [82] and Doll and Torkzadeh [83]. Empirical studies provide robust support for the pivotal role of perceived usefulness in shaping adoption behavior. For example, Pikkarainen et al. [84] demonstrated that perceived usefulness is a key driver of users’ willingness to adopt innovative and user-friendly technologies, as it provides greater flexibility and efficiency. Similarly, Hanafizadeh et al. [85] emphasized that an individual’s intention to use a specific IS largely depends on their perception of its utility in task performance. In the taxation context, Lymer et al. [86] further highlighted that perceived usefulness exerts a significant and positive influence on taxpayers’ willingness to adopt e-Filing services. Based on this body of evidence, the following hypothesis is advanced:

H6. 

Perceived usefulness has a positive effect on intention to use.

3.6. Satisfaction

User satisfaction refers to the degree of users’ contentment with an IS, arising when the services and support provided by the system meet or exceed user expectations [17]. In the revised D&M model, user satisfaction is conceptualized as a key outcome of the perceived benefits of the system, which subsequently strengthens both satisfaction levels and the intention to continue using the system [87]. In the taxation domain, empirical studies reinforce this perspective. For instance, Puthur et al. [88] and Maharani et al. [89] demonstrated that satisfaction significantly and positively influences taxpayers’ willingness to adopt e-Filing services, with observable effects on actual system use. Similarly, Hambali [90] identified a direct correlation between user satisfaction and the effective utilization of e-Filing systems. Accordingly, this study posits that user satisfaction exerts a favorable influence on both intentions to use and actual usage. The hypotheses are therefore formulated as follows:

H7. 

Satisfaction has a positive effect on intention to use.

H8. 

Satisfaction has a positive effect on actual use.

3.7. Intention to Use

In both the TAM and the D&M model, intention to use is considered a focal variable, as it reflects the likelihood that an individual will adopt a given technology [17,19]. To operationalize this construct, prior studies often employ indicators such as dependency and future usage expectations [91,92]. These indicators serve as critical precursors to actual adoption behavior, providing the attitudinal foundation through which technology usage materializes. Empirical evidence supports this linkage. For instance, Tao [93] demonstrated a direct correlation between intention and actual use, confirming the predictive power of intention in technology adoption research. In the taxation domain, similar findings have been reported. Hernandez-Ortega [94], examining e-invoicing adoption among 100 early adopter firms in Spain, found that trust cultivated through the system significantly increased intention to adopt, which in turn fostered sustained usage. Likewise, Ramdhony et al. [95], drawing on a sample of 315 e-filing users in Mauritius, identified user satisfaction as a key determinant of continued intention to use and, consequently, actual utilization of online tax-filing services. Building on these insights, this study posits the following hypothesis:

H9. 

Intention to use has a positive effect on actual use.

3.8. Actual Use

Unlike intention to use, actual use reflects users’ behaviors rather than their attitudes toward an IS [17]. In most empirical studies, actual use is operationalized through self-reported measures rather than objective indicators such as system-generated usage logs [96]. According to Turner [97], subjective measures of usage typically capture the frequency and intensity with which individuals report engaging with a given technology. Empirical evidence in the taxation context suggests that the actual use of technologies designed to facilitate compliance can significantly reduce compliance costs. For instance, Lee [98], reviewing the Republic of Korea’s experience with mandatory electronic tax invoices since their introduction in 2011, highlighted their role in reducing compliance costs while simultaneously enhancing the transparency of business transactions. Similarly, Saptono et al. [26], examining the implementation of e-Filing and e-Form systems in Indonesia, found that actual use of these platforms contributed to lowering taxpayers’ compliance burdens. Building on these insights, this study proposes the following hypothesis:

H10. 

Actual use has a positive effect on perceived reduced compliance costs.

3.9. Perceived Reduced Compliance Costs and Tax Compliance Intention

Perceived reduced compliance costs refer to taxpayers’ subjective evaluation of the extent to which administrative obligations—such as time, monetary expenses, and procedural burdens—are minimized through a given system [99]. Meanwhile, tax compliance intention reflects taxpayers’ willingness and planned effort to fulfil their obligations, typically assessed through completeness, timeliness, accuracy, and prioritization in reporting [100,101]. While objective measures of compliance costs and tax compliance behavior are often considered superior, both constructs can also be reliably captured through subjective perceptions, as taxpayers’ decisions are largely shaped by their cognitive evaluations of burden and obligation [102,103]. Empirical studies further demonstrate that perceptions of reduced compliance costs strengthen compliance intentions; for instance, Palil and Mustapha [104] and Musimenta [105] provide evidence that simplification and diminished compliance burdens significantly enhance taxpayers’ willingness to comply. Accordingly, this study proposes the following hypothesis:

H11. 

Perceived reduced compliance costs have a positive effect on tax compliance intention.

4. Materials and Methods

4.1. Questionnaire Development

We employed a closed-response questionnaire as the measuring instrument for our research. This format was selected because it enables the calculation of average scores for each item, which is a standard methodological approach for analyzing multi-item constructs [106,107].

The questionnaire consisted of two main sections. The first section collected respondents’ demographic and background information, including gender, age, education, professional experience, and level of familiarity with CTAS. The second section measured the core variables of the study. These constructs were operationalized using multiple items adapted from established theoretical indicators discussed in the previous section and from validated measures commonly used in IS success research.

All constructs in this study were modeled reflectively, meaning that variations in the latent variable are assumed to be manifested consistently across their observed items. The full list of items employed in this study is presented in

Table 1. The Research Instrument.

Variable	No.	Indicator	Item	Statement	Source	Factor Loading	Cronbach’s Alpha
System quality	1	User-friendly	SysQ1	The CTAS interface is easy to understand.	[108]	0.799	0.753
	2	Easy to use	SysQ2	CTAS is easy to use for tax compliance.	[109]	0.874
	3	Performance reliability	SysQ3	CTAS is technically reliable.	[110]	0.780
Service quality	4	Readiness for service	SrvQ1	CTAS is reliably available.	[111]	0.898	0.910
	5	Safe transactions	SrvQ2	CTAS ensures secure and confidential transactions.	[112]	0.842
	6	Availability	SrvQ3	CTAS is accessible anytime.	[113]	0.884
	7	Individual attention	SrvQ4	CTAS services are user oriented.	[114]	0.797
	8	Specific needs for users	SrvQ5	CTAS features match specific tax compliance needs.	[115]	0.862
Information quality	9	Precise information	InfQ1	CTAS provides accurate information	[26]	0.890	0.947
	10	Up-to-date information	InfQ2	Information in CTAS is regularly updated according to the latest regulations.	[67]	0.850
	11	Sufficient information	InfQ3	CTAS provides sufficient information to complete tax compliance obligations.	[116]	0.951
	12	Reliable information	InfQ4	Information in CTAS is trustworthy and consistent.	[117]	0.950
	13	Useful information	InfQ5	Information in CTAS is useful and relevant to my tax compliance needs.	[118]	0.899
Perceived ease of use	14	Ease of becoming skillful	PEU1	I can quickly become proficient in using CTAS.	[119]	0.916	0.888
	15	Ease of remembering	PEU2	I can easily remember how to perform tasks when using CTAS.	[120]	0.893
	16	Ease of interaction	PEU3	Interacting with CTAS is effortless	[121]	0.900
Perceived usefulness	17	Perceived effectiveness	PU1	CTAS makes my tax obligations more effective and structured.	[122]	0.923	0.924
	18	Performance quality	PU2	CTAS improves accuracy and quality in tax administration.	[123]	0.943
	19	Productivity	PU3	CTAS increases my productivity in tax-related work.	[124]	0.929
Satisfaction	20	Satisfaction with the system	SF1	Overall, I am satisfied with using CTAS.	[125]	0.959	0.913
	21	Expectations	SF2	CTAS meets my expectations in fulfilling tax obligations	[126]	0.959
Intention to use	22	Dependency	IU1	I rely on CTAS to fulfil my tax obligations.	[127]	0.771	0.851
	23	Tendency to use	IU2	I intend to continue using CTAS in the future.	[128]	0.909
	24	Duration of future use	IU3	I will use CTAS frequently for my work.	[129]	0.950
Actual use	25	Frequency	AU1	I use CTAS routinely for tax administration activities.	[130]	0.839	0.701
	26	Intensity	AU2	I maximize CTAS according to my tax administration needs.	[131]	0.911
Perceived reduced compliance cost	27	Compliance time	RCC1	CTAS makes tax administration faster than visiting the tax office.	[132]	0.916	0.879
	28	Financial cost	RCC2	CTAS is more cost-efficient than visiting the tax office.	[105]	0.958
	29	Administrative efficiency	RCC3	CTAS reduces the need for paper documents in tax administration.	[133]	0.813
Tax compliance intention	30	Completeness	TCI1	Filing SPT through CTAS ensures the complete disclosure of all tax obligations.	[134]	0.931	0.867
	31	Timeliness	TCI2	CTAS helps me submit tax returns on time.	[135]	0.801
	32	Accuracy	TCI3	CTAS ensures accurate tax payment and prevents penalties.	[136]	0.919
	33	Prioritization	TCI4	I prioritize paying taxes through CTAS over other financial obligations.	[137]	0.725

.

A six-point Likert scale, ranging from 1 (“strongly disagree”) to 6 (“strongly agree”), was employed to evaluate each item. The use of an even-numbered scale was deliberately selected, as prior studies have argued that odd-point scales may be suboptimal and provide limited methodological benefits [138,139]. One key concern is the midpoint option in odd-point scales, which may lengthen response time and introduce ambiguity into item interpretation [140].

In contrast, even-point scales compel respondents to take a definitive stance—either agreement or disagreement—rather than defaulting to a neutral position. They also help reduce potential misinterpretation of the midpoint. Prior evidence indicates that some respondents tend to choose the midpoint simply to avoid expressing negative opinions. In many cases, this response may reflect mild disagreement [141]. These considerations support the use of a six-point scale in this study.

Prior to the main survey, a pilot study was conducted with 30 respondents, as recommended by Bujang et al. [142]. The objective was to assess the validity and reliability of the questionnaire (see Table 1).

Following Krishnaswamy et al. [143], both content and construct validity were assessed. Content validity was established during the instrument development stage through literature review and expert evaluation. One academic expert and two industry practitioners reviewed the questionnaire and confirmed the clarity and relevance of all items.

Construct validity was examined through exploratory factor analysis (EFA). The results demonstrated that all items significantly loaded on their intended constructs, with standardized factor loadings exceeding the recommended threshold of 0.7 [144].

Reliability was assessed using Cronbach’s alpha. All constructs achieving values above 0.70, indicating satisfactory internal consistency [145]. Since no items required elimination or substantial revision, the instrument was deemed valid and reliable in its entirety and retained for use in the main study.

4.2. Sample and Data Collection Procedures

This study employed a cross-sectional design, in which data were collected from the target population at a single point in time [146], specifically during 2025, the first year of CTAS implementation. The population of interest did not consist of ordinary taxpayers, but rather tax professionals, namely certified tax consultants and tax accountants [20]. According to Minister of Finance Regulation No. 111/PMK.03/2014, tax consultants are individuals who provide advisory services to both individual and corporate taxpayers regarding the exercise of tax rights and obligations in accordance with prevailing regulations. Their professional role also includes assisting or representing clients in dealings with the tax authority. To obtain a licence to practise, tax consultants are required to pass the tax consultant certification examination and earn an official certificate, which is categorized into levels A, B, and C. Specifically, holders of level A are authorized to serve individual taxpayers, holders of level B are authorized to serve corporate taxpayers, while holders of level C are authorized to serve taxpayers with international scope. All certified consultants are formally registered in the Ministry of Finance’s Tax Consultant Information System (called ‘SIKOP’ in Indonesia; https://sikop.kemenkeu.go.id/front/carikonsultan (accessed on 30 January 2026)). At the time of this study, there were 7772 certified and active tax consultants listed in SIKOP.

In contrast, tax accountants are professionals primarily responsible for recording financial transactions, analyzing and calculating tax liabilities, and ensuring their companies’ compliance with prevailing tax regulations [147]. Although many tax accountants hold relevant academic qualifications and have completed specialized tax training program such as the Brevet, they typically do not possess formal certification equivalent to that of tax consultants and, consequently, are not listed in SIKOP. As a result, no reliable data are available regarding the precise population size of tax accountants in Indonesia.

To collect the data, we employed a convenience sampling method, selected for its efficiency, practicality, and cost-effectiveness, despite its inherent limitation in terms of generalizability [146]. The questionnaire was designed using Google Forms and disseminated online through multiple channels, including the WhatsApp group of the Indonesian Tax Consultants Association and LinkedIn. In addition, the first author leveraged a recurring weekly webinar series (held every Wednesday via Zoom), which he regularly hosts. These webinars, typically attended by tax accountants seeking guidance on practical taxation matters, provided a particularly effective and contextually relevant avenue for reaching potential respondents. Data collection commenced on 3 September 2025 and continued until the target sample size was achieved.

The determination of the minimum sample size followed the widely applied rule of thumb that recommends ten respondents per questionnaire item [148], yielding a target of 330 respondents for the 33-item survey instrument (see Table 1). This approach was deemed appropriate given the unknown size of the broader tax accountant population, for which alternative rules based on population percentages would have been less relevant [149]. By the end of the collection period on 1 October 2025, a total of 401 responses had been gathered. This represents a high response rate within a relatively short timeframe of approximately one month, likely facilitated by the provision of a modest incentive of IDR 25,000 offered to 200 randomly selected participants—a common strategy in survey research to boost participation rates [150]. Following a non-engagement bias analysis, 47 responses were excluded because the respondents reported little to no familiarity with CTAS. The final dataset therefore comprised 354 valid responses (88.28%), which exceeded the minimum target of 330.

The demographic and professional characteristics of the 354 respondents are presented in

Table 2. Respondent Profile.

	Category	Frequency	Percentage (%)
Gender	Male	185	52.26
	Female	169	47.74
Age	Generation Z (18–28 years)	55	15.54
	Millennials (29–44 years)	171	48.31
	Generation X (45–59 years)	118	33.33
	Baby Boomers (≥60 years)	10	2.82
Education	3-year Diploma	30	8.47
	4-year Diploma	5	1.41
	Bachelor	257	72.60
	Master	54	15.25
	PhD	8	2.26
Professional Role	Tax Consultant	84	23.73
	Tax Accountant	270	76.27
Certificate Level	A	44	12.43
	B	26	7.34
	C	14	3.95
	Brevet	193	54.52
	None	77	21.75
Working Experience	Fresh Graduate (0–<1 years)	22	6.21
	Junior (1–<3 years)	44	12.43
	Intermediate (3–<5 years)	52	14.69
	Senior (5–<10 years)	105	29.66
	Expert (>10 years)	131	37.01
CTAS Training	Yes	312	88.14
	No	42	11.86
CTAS Familiarity	Familiar	108	30.51
	Highly Familiar	246	69.49

. Gender composition is nearly even, with men (52.26%) only slightly outnumbering women (47.74%). In terms of age, millennials form the largest cohort (48.31%), followed by generation X (33.33%), indicating that most respondents are in the mature and professionally active age bracket, while younger Generation Z (15.54%) and Baby Boomers (2.82%) provide supplementary generational diversity. Educational attainment is notably high, with nearly three-quarters holding a bachelor’s degree (72.60%), complemented by postgraduate qualifications (17.51%). Professional roles are dominated by tax accountants (76.27%), yet a considerable proportion are consultants (23.73%). Certification levels further validate professional credibility, with only about one-fifth of respondents (21.75%) lacking any form of certification. Work experience skews toward senior and expert levels (66.67%), underscoring depth of practice. Importantly, the vast majority have attended CTAS training (88.14%) and report high familiarity (69.49%), suggesting respondents are both informed and directly relevant to evaluating the system.

4.3. Data Analysis Technique

In the present research, Structural Equation Modeling (SEM) was adopted as the principal analytical approach because it provides a comprehensive framework to examine interdependent relationships among multiple constructs. Unlike conventional regression techniques that generally assume precise measurement, SEM explicitly integrates measurement error into the model, thereby producing more reliable parameter estimates. Its methodological strength has made SEM increasingly prominent in social sciences, as it allows scholars to simultaneously test both measurement and structural models in a single analysis [151]. This capability is especially relevant to the current study, which operationalizes abstract latent variables through several observed indicators. Since these indicators cannot perfectly mirror the underlying concepts, relying solely on regression analysis would risk oversimplification and potentially distort statistical inference.

Beyond mitigating measurement error, SEM provides significant potential for advancing theoretical development [152]. Cheng [153] further argues that while multiple regression is effective for predicting outcomes based on predictor variables, it falls short in proposing and validating new theoretical linkages derived from empirical data. In contrast, SEM facilitates iterative estimation, model reassessment, and the exploration of theoretically meaningful pathways. This capability is particularly valuable in the context of our research, which integrates the TAM and the D&M model to evaluate CTAS. As this domain remains relatively underexplored in the literature, the application of SEM allows us to introduce and empirically validate novel linkages, thereby offering fresh insights that contribute both theoretically and practically to the understanding of IS in taxation.

In SEM, two dominant approaches are commonly employed: Partial Least Squares SEM (PLS-SEM) and Covariance-Based SEM (CB-SEM) [152]. Each method presents distinct advantages depending on the research objectives and contextual conditions, and the choice between them requires careful methodological consideration. For the purposes of this study, PLS-SEM was selected for two compelling reasons.

First, PLS-SEM is particularly well suited for datasets that deviate from normality, as it is explicitly designed to provide robust estimates of latent constructs under such conditions [154]. According to Hair et al. [155], non-normality is indicated when skewness and kurtosis values exceed the absolute threshold of ±1. In our dataset, while most skewness values fell within the acceptable range (−0.837 to 1.229), one indicator slightly exceeded this limit. More critically, several indicators displayed excess kurtosis above +1, with values ranging from 1.042 to 1.874, suggesting heavier-tailed distributions. These deviations, particularly in kurtosis, confirm that the data cannot be considered normally distributed. Accordingly, the use of PLS-SEM is methodologically justified in this study. Second, this study involved a relatively modest sample of 354 respondents, which is small compared to the broader population of 7772 certified tax consultants and an even larger, but undocumented, group of tax accountants in Indonesia. Prior research has demonstrated that PLS-SEM consistently produces reliable parameter estimates even with limited sample sizes, making it more appropriate than CB-SEM in contexts where the available data are constrained [154]. We used SmartPLS 4.1.1.2 software to implement the PLS-SEM technique in our analysis.

5. Results

In reporting the results of our PLS-SEM analysis, we adhered to the systematic procedures outlined by Hair et al. [156]. The first step involved evaluating the measurement model, with all constructs specified as reflective. This stage assessed the reliability and validity of the indicators to ensure they accurately reflected their underlying latent constructs. The second step comprised the analysis of descriptive statistics, offering insights into data distribution, central tendency, and variability to capture overall response patterns. The third step entailed evaluating the structural model, which assessed the model’s explanatory power and tested the hypothesized relationships using path coefficients. Finally, as part of the robustness assessment, we examined potential nonlinear effects, endogeneity, and heterogeneity to ensure that the estimated results were both theoretically coherent and empirically robust.

5.1. Measurement Model Evaluation

Table 3. Validity and Reliability of Measurement Items.

Variable	Item	Standardized Loading	AVE	Cronbach’s Alpha	CR
System Quality	SysQ1	0.850	0.646	0.723	0.845
	SysQ2	0.825
	SysQ3	0.731
Service Quality	SrvQ1	0.773	0.568	0.812	0.868
	SrvQ2	0.706
	SrvQ3	0.761
	SrvQ4	0.767
	SrvQ5	0.760
Information Quality	InfQ1	0.772	0.649	0.864	0.902
	InfQ2	0.758
	InfQ3	0.866
	InfQ4	0.789
	InfQ5	0.838
Perceived Ease of Use	PEU1	0.847	0.696	0.787	0.873
	PEU2	0.817
	PEU3	0.839
Perceived Usefulness	PU1	0.893	0.767	0.848	0.908
	PU2	0.884
	PU3	0.849
Satisfaction	SF1	0.947	0.896	0.884	0.945
	SF2	0.947
Intention to Use	IU1	0.679	0.720	0.798	0.883
	IU2	0.920
	IU3	0.924
Actual Use	AU1	0.867	0.784	0.726	0.879
	AU2	0.903
Perceived Reduced Compliance Cost	RCC1	0.850	0.740	0.824	0.895
	RCC2	0.898
	RCC3	0.831
Tax Compliance Intention	TCI1	0.794	0.628	0.802	0.871
	TCI2	0.767
	TCI3	0.830
	TCI4	0.778

Note: Values in italics denote indicators that do not meet the prescribed threshold.

presents the results of the measurement model assessment. As shown, almost all standardized factor loadings exceeded the recommended threshold of 0.70 [155], indicating that each indicator was strongly correlated with its corresponding latent construct. This finding underscores the effectiveness of the indicators in reflecting their underlying conceptual domains. However, this was not the case for IU1 (dependency), which exhibited a loading value below the recommended threshold. This may be attributed to the fact that dependency reflects a post-adoption behavioral state rather than a deliberate intention [157], making it less conceptually aligned with the construct of intention to use. Therefore, this indicator was excluded from subsequent analyses. Furthermore, the Average Variance Extracted (AVE) values for all constructs exceeded the critical threshold of 0.50, confirming the presence of convergent validity [156]. Collectively, these findings indicate that the remaining indicators adequately represent their corresponding latent constructs.

The internal consistency of the measurement model was assessed using Cronbach’s alpha, which remains the most widely applied reliability indicator in social science research [158]. However, as noted by DeVellis [159], Cronbach’s alpha alone may not fully capture the reliability of a scale, as it does not ensure unidimensionality. Moreover, the statistic assumes a restrictive tau-equivalent model, in which all factor loadings and error variances are constrained to be equal [160]. To address these limitations, we additionally computed composite reliability (CR) to evaluate construct reliability more rigorously. Unlike Cronbach’s alpha, CR accounts for the individual factor loadings of items and does not treat all indicators as equally reliable [158]. As reported in Table 3, both reliability indices demonstrated satisfactory values across all constructs, with coefficients lying between 0.70 and 0.95, confirming high internal consistency and construct reliability [161,162].

Beyond the preceding assessments, we further verified that our model satisfied the requirements of discriminant validity. Establishing discriminant validity is crucial to ensure that each latent construct captures a unique dimension of the theoretical framework and is not excessively correlated with other constructs in the model [155]. This step safeguards the conceptual integrity of the model by confirming that the constructs provide distinct explanatory power rather than overlapping conceptually.

To evaluate discriminant validity, we applied both the Fornell–Larcker criterion and the Heterotrait–Monotrait (HTMT) ratio, as these complementary methods offer a comprehensive perspective on construct distinctiveness. The results, presented in

Table 4. The Square Root of AVE (at Diagonal) and Correlation Coefficient.

	SysQ	SrvQ	InfQ	PEU	PU	SF	IU	AU	RCC	TCI
SysQ	0.804
SrvQ	0.627	0.754
InfQ	0.532	0.710	0.805
PEU	0.551	0.633	0.599	0.834
PU	0.532	0.668	0.675	0.697	0.876
SF	0.549	0.643	0.564	0.616	0.671	0.947
IU	0.365	0.499	0.514	0.473	0.579	0.566	0.849
AU	0.185	0.354	0.375	0.331	0.398	0.319	0.492	0.885
RCC	0.370	0.438	0.476	0.405	0.455	0.385	0.468	0.444	0.860
TCI	0.340	0.467	0.498	0.382	0.493	0.390	0.535	0.504	0.653	0.792

and

Table 5. HTMT.

	SysQ	SrvQ	InfQ	PEU	PU	SF	IU	AU	RCC
SrvQ	0.811
InfQ	0.860	0.860
PEU	0.714	0.768	0.713
PU	0.680	0.801	0.787	0.822
SF	0.685	0.744	0.644	0.707	0.776
IU	0.428	0.568	0.553	0.524	0.631	0.604
AU	0.251	0.466	0.462	0.433	0.496	0.394	0.588
RCC	0.486	0.550	0.566	0.504	0.550	0.457	0.533	0.570
TCI	0.449	0.584	0.597	0.472	0.601	0.463	0.604	0.662	0.794

, indicate that the square root of the AVE for each latent variable exceeds its correlations with other constructs, thereby fulfilling the Fornell–Larcker requirement and confirming that each construct explains more variance in its own indicators than in those of other constructs [163]. Furthermore, all HTMT ratios were found to be below the recommended threshold of 0.90 [164], providing additional support that the constructs are empirically distinct from one another. Taken together, these results provide strong evidence of discriminant validity, indicating that each construct is uniquely and accurately represented by its corresponding indicators within the model.

5.2. Descriptive Statistics

Table 6. Descriptive Statistics of Measurement Items.

Variable	Item	Min.	Max.	Mean	Std. Deviation
System Quality	SysQ1	1	6	3.927	1.058
	SysQ2	1	6	3.901	1.193
	SysQ3	1	6	2.037	1.203
Service Quality	SrvQ1	1	6	3.342	1.388
	SrvQ2	1	6	3.884	1.184
	SrvQ3	1	6	3.582	1.420
	SrvQ4	1	6	3.873	1.211
	SrvQ5	1	6	3.907	1.162
Information Quality	InfQ1	1	6	3.718	1.216
	InfQ2	1	6	3.983	1.193
	InfQ3	1	6	3.850	1.194
	InfQ4	1	6	3.986	1.166
	InfQ5	1	6	4.025	1.095
Perceived Ease of Use	PEU1	1	6	4.028	1.122
	PEU2	1	6	4.051	1.043
	PEU3	1	6	3.565	1.409
Perceived Usefulness	PU1	1	6	3.915	1.230
	PU2	1	6	3.935	1.214
	PU3	1	6	3.602	1.356
Satisfaction	SF1	1	6	3.655	1.357
	SF2	1	6	3.667	1.385
Intention to Use	IU2	1	6	4.184	1.083
	IU3	1	6	4.150	1.064
Actual Use	AU1	1	6	4.325	0.905
	AU2	1	6	4.192	0.981
Perceived Reduced Compliance Cost	RCC1	1	6	4.218	1.198
	RCC2	1	6	4.390	1.066
	RCC3	1	6	4.333	1.074
Tax Compliance Intention	TCI1	1	6	4.184	1.032
	TCI2	1	6	4.198	1.079
	TCI3	1	6	4.316	0.978
	TCI4	1	6	4.333	1.192

reports the mean and standard deviation for all Likert-scale questionnaire items. The mean values summarize the central tendency of responses, while standard deviations indicate the extent to which these means represent data variability [165]. As previously described, a six-point even scale (1 = strongly disagree, 6 = strongly agree) was employed to avoid a neutral midpoint and to facilitate the dichotomization of responses into positive and negative categories [166]. In this context, a threshold of 4 out of 6 was used, where mean scores below this threshold indicate a negative sentiment toward the item and those above indicate a positive sentiment. This cutoff was set slightly above the conceptual midpoint (3.5) to account for the tendency of respondents to lean toward less favorable evaluations when an even-point scale is used [141].

The findings presented in Table 6 reveal that tax professionals generally hold negative perceptions of system quality (mean values ranging from 2.037 to 3.927), service quality (mean values ranging from 3.342 and 3.907), and information quality (mean values ranging from 3.718 and 3.986) in relation to the CTAS platform. The only indicator receiving a positive response was InfQ5 (mean value = 4.025), which assesses information quality in terms of its usefulness. These results contrast with those reported by Saptono et al. [26], who examined Indonesia’s e-Filing and e-Form services and concluded that tax consultants hold a highly positive perception of all three quality dimensions in the D&M model. This discrepancy suggests that, unlike the relatively mature and well-optimized e-Filing and e-Form systems, the implementation of CTAS remains in its early stage of adoption, where technical limitations, usability challenges, and adaptation issues may have constrained users’ overall perceptions of quality.

Consistent with this interpretation, the perceived ease of use dimension also yielded mixed responses, with the item “interacting with CTAS is easy” (PEU3, mean value = 3.565) receiving the lowest score despite two other indicators reflecting generally positive sentiments. Similarly, the perceived usefulness construct was viewed unfavorably (mean values ranging from 3.602 to 3.915), particularly regarding the extent to which CTAS enhances productivity in handling tax-related tasks (PU3). Taken together, these patterns corroborate the notion that users’ negative evaluations of system quality and perceived utility are intertwined with broader usability and adaptation challenges. Such issues appear to culminate in the relatively low level of user satisfaction observed for CTAS (mean values ranging from 3.655 to 3.667), underscoring the need for further refinement of the system’s functionality, user interface, and support mechanisms.

Nevertheless, our respondents—comprising tax consultants and tax accountants—appear to remain proactive in adapting to this newly launched digital tax administration system. This is reflected in the relatively high mean values for intention to use (ranging from 4.150 to 4.184). Likewise, actual system uses also received positive responses (mean values ranging from 4.192 to 4.325), which is unsurprising given that fulfilling tax obligations through CTAS constitutes a core component of their professional responsibilities, both for clients and within their employing organizations.

Moreover, a glimpse of optimism regarding the successful implementation of CTAS emerges from the respondents’ positive perceptions of reduced compliance costs (mean values ranging from 4.218 to 4.390) and tax compliance intention (mean values ranging from 4.184 to 4.333). These findings indicate that, despite encountering technical and usability challenges, tax professionals generally recognize the system’s potential to streamline compliance procedures and enhance voluntary tax adherence. Additionally, since the standard deviations across all questionnaire items were considerably lower than their respective means, it can be inferred that most responses clustered closely around the average values [165]. Accordingly, the summarized data provide a representative and stable depiction of the overall dataset.

5.3. Structural Model Evaluation

Before reporting the results of hypothesis testing, it is essential to evaluate the quality of the structural model through a series of statistical assessments, as shown in

Table 7. Model Quality Criteria.

Relationship	VIF	${\mathit{Q}}^2$	R-Square	Assessment	${\mathit{f}}^2$	Assessment
SysQ → SF	1.690	0.449	0.464	Moderate	0.051	Small
SrvQ → SF	2.445				0.110	Small
InfQ → SF	2.070				0.029	Small
PEU → PU	1.000	0.480	0.485	Moderate	0.943	Large
PEU → IU	2.105	0.257	0.363	Moderate	0.002	No Effect
PU → IU	2.381				0.068	Small
SF → IU	1.972				0.070	Small
SF → AU	1.418	0.114	0.235	Weak	0.006	No Effect
IU → AU	1.418				0.174	Moderate
AU → RCC	1.000	0.090	0.197	Weak	0.245	Moderate
RCC → TCI	1.000	0.060	0.427	Moderate	0.744	Large

. First, multicollinearity among the predictor constructs was examined using the Variance Inflation Factor (VIF). The results indicate that all VIF values were well below the recommended threshold of 3.3 [167], suggesting that the predictors were not excessively correlated. Maintaining low levels of multicollinearity is critical to preserving the integrity of the structural relationships under investigation and ensuring that the estimated results are not biased by redundant or overlapping information [168]. In addition, the model’s predictive relevance was assessed through the blindfolding procedure, which calculates the Stone–Geisser $Q^2$ statistic [169,170]. The results yielded $Q^2$ values greater than zero, confirming that the model possesses adequate explanatory power for the endogenous constructs [155].

To further assess the explanatory power of the independent variables in predicting variance across the endogenous constructs, we examined the R-square values [171,172]. As a general guideline, R-square values of 0.75, 0.50, and 0.25 are interpreted as substantial, moderate, and weak, respectively [173,174]. According to Table 7, the determinants of satisfaction, perceived usefulness, intention to use, and tax compliance yielded R-square values of 0.464, 0.485, 0.363, and 0.427, respectively. These results suggest that the model accounts for approximately 46% of the variance in CTAS user satisfaction, 49% in perceived usefulness, 36% in intention to use, and 43% in tax compliance intention. Overall, these values suggest moderate Overall, these values suggest moderate explanatory power for the main outcome constructs, although a substantial portion of variance remains unexplained. In addition, several intermediate endogenous constructs exhibit weaker explanatory strength, indicating that additional contextual, organizational, or behavioral factors not included in the model may also influence these outcomes. Nonetheless, such modest R-square values are generally acceptable in behavioral research, where attitudes, perceptions, and intentions are inherently complex and influenced by numerous unobserved factors [175].

In addition to evaluating the predictive power of the model, researchers are encouraged to assess how the removal of specific predictor constructs influences the effect size ($f^2$) of each predictor on its corresponding endogenous variable. According to Cohen’s [176] guidelines, $f^2$ values of 0.02, 0.15, and 0.35 represent small, medium, and large effects, respectively, whereas values below 0.02 indicate negligible effects. As reported in Table 7, most relationships within the model exhibit small to moderate effect sizes. Notably, two relationships—between perceived ease of use and perceived usefulness ($f^2$ = 0.943), and between perceived reduced compliance costs and tax compliance intention ($f^2$ = 0.744)—demonstrate large effects, underscoring their pivotal roles in the structural framework. Conversely, the effects of perceived ease of use on intention to use ($f^2$ = 0.002) and satisfaction on actual use ($f^2$ = 0.006) are negligible. Overall, these findings indicate that while most relationships exert moderate influence, the two key constructs—perceived ease of use and perceived reduced compliance costs—serve as primary levers in explaining system adoption and compliance behavior toward CTAS.

5.4. Path Analysis

In this study, path analysis was conducted using a bootstrapping procedure with 5000 iterations, following the recommendations of Hair et al. [155], to test the significance of the hypothesized relationships. This analysis examined both the direct and indirect relationships among latent constructs within the structural model, allowing for a comprehensive understanding of how the proposed factors interact to shape behavioral and compliance outcomes.

5.4.1. Direct Effect

Table 8. Path Coefficients and Significances.

Hypothesis	Relationship	$\mathit{\beta }$	Std. Deviation	t-Statistics	Decision
H1	SysQ → SF	0.214	0.057	3.782 **	Accepted
H2	SrvQ → SF	0.380	0.069	5.543 **	Accepted
H3	InfQ → SF	0.180	0.063	2.861 **	Accepted
H4	PEU → IU	0.051	0.064	0.802	Rejected
H5	PEU → PU	0.697	0.031	22.546 **	Accepted
H6	PU → IU	0.321	0.071	4.546 **	Accepted
H7	SF → IU	0.296	0.070	4.231 **	Accepted
H8	SF → AU	0.083	0.064	1.306	Rejected
H9	IU → AU	0.434	0.070	6.167 **	Accepted
H10	AU → RCC	0.444	0.070	6.369 **	Accepted
H11	RCC → TCI	0.653	0.048	13.678 **	Accepted

Note: ** represents statistical significance at the 1% level.

and Figure 5 present the SmartPLS outputs illustrating the direct effects of the predictor constructs on the endogenous variables. The results show that all factors positively influencing user satisfaction were statistically significant at the 1% level, thereby supporting H1 to H3. Among these predictors, service quality ($\beta$ = 0.380, p-value < 0.01) exhibited the strongest effect, followed by system quality ($\beta$ = 0.214, p-value < 0.01) and information quality ($\beta$ = 0.180, p-value < 0.01). These results suggest that the technical, semantic, and service-related dimensions of CTAS play a central role in shaping users’ satisfaction with the system.

With respect to intention to use, the analysis shows that perceived usefulness ($\beta$ = 0.321, p-value < 0.01) and user satisfaction ($\beta$ = 0.296, p-value < 0.01) significantly and positively influence users’ behavioral intention, providing strong support for H6 and H7. In contrast, the direct effect of perceived ease of use on intention to use ($\beta$ = 0.051, p-value > 0.05) was found to be statistically insignificant, leading to the rejection of H4. This finding implies that users’ willingness to adopt CTAS is more strongly driven by its perceived utility and their satisfaction with it, rather than by perceptions of its ease of operation. Nevertheless, perceived ease of use was found to have a substantial and highly significant positive effect on perceived usefulness ($\beta$ = 0.697, p-value < 0.01), thereby supporting H5 and confirming a core proposition of the TAM—that systems perceived as easier to use are also perceived as more useful.

In terms of actual use, intention to use ($\beta$ = 0.434, p-value < 0.01) was identified as a significant and strong predictor, supporting H9, whereas the direct influence of satisfaction ($\beta$ = 0.083, p-value > 0.05) was not significant, resulting in the rejection of H8. This indicates that actual engagement with CTAS is primarily driven by behavioral intention rather than affective satisfaction. Finally, post-adoption outcomes provide robust empirical support for the extended model. Actual use demonstrated a strong positive effect on perceived reduced compliance costs ($\beta$ = 0.444, p-value < 0.01), thus supporting H10, suggesting that users who actively interact with CTAS tend to perceive it as more efficient and less burdensome in managing tax obligations. Furthermore, perceived reduced compliance costs exerted a substantial positive influence on tax compliance intention ($\beta$ = 0.653, p-value < 0.01), confirming H11. This relationship highlights the pivotal role of perceived cost reduction in fostering compliance behavior through enhanced system adoption and utilization.

5.4.2. Indirect Effect

Table 9. Specific Indirect Effect.

D&M Model
Relationship	$\mathit{\beta }$	Std. Deviation	t-Statistics	Inference
SysQ → SF → AU	0.018	0.015	1.222	IU is a full mediator
SysQ → SF → IU → AU	0.028	0.010	2.779 **
SrvQ → SF → AU	0.032	0.026	1.236	IU is a full mediator
SrvQ → SF → IU → AU	0.049	0.015	3.153 **
InfQ → SF → AU	0.015	0.014	1.107	IU is a full mediator
InfQ → SF → IU → AU	0.023	0.011	2.190 **
TAM
Relationship	$\mathit{\beta }$	Std. Deviation	t-Statistics	Inference
PEU → IU → AU	0.022	0.029	0.764	PU is a full mediator
PEU → PU → IU → AU	0.097	0.029	3.317 **
Benefits from Actual Use of CTAS
Relationship	$\mathit{\beta }$	Std. Deviation	t-Statistics	Inference
AU → RCC → TCI	0.290	0.058	4.998 **	Significant

Note: ** represents statistical significance at the 1% level.

presents the results of the indirect effect analysis for the structural model developed under the D&M and TAM frameworks. The findings reveal that, within the D&M model, the indirect paths operating solely through satisfaction—specifically, the effects of system quality on actual use ($\beta$ = 0.018, p-value > 0.05), service quality on actual use ($\beta$ = 0.032, p-value > 0.05), and information quality on actual use ($\beta$ = 0.015, p-value > 0.05)—are not statistically significant. However, when intention to use is incorporated as a sequential mediator following satisfaction, these relationships become significant at the 1% level ($\beta$ = 0.028, 0.049, dan 0.023, respectively). This shift from insignificance to significance highlights that intention to use—not satisfaction alone—serves as the primary mediating mechanism that transforms users’ perceptions of system, service, and information quality into actual behavioral engagement with CTAS. In other words, while satisfaction captures users’ evaluative judgments, it is the formation of a strong behavioral intention that ultimately drives them to actively use the system.

However, within the TAM framework, perceived usefulness emerges as the pivotal mediating construct that links users’ perceptions of system ease to their actual engagement with CTAS. The indirect pathway operating solely through intention to use—specifically, the effect of perceived ease of use on actual use ($\beta$ = 0.022, p-value > 0.05)—is statistically insignificant, indicating that ease of use alone does not directly translate into behavioral engagement. However, when perceived usefulness is introduced as an additional mediator in sequence, the indirect effect becomes statistically significant at the 1% level ($\beta$ = 0.097, p-value < 0.01). This shift underscores that perceived usefulness serves as a full mediator, meaning that users’ perceptions of system ease enhance adoption behavior primarily by strengthening their belief in the system’s practical utility rather than through direct motivational intentions.

Furthermore, regarding the practical benefits of CTAS adoption, the analysis shows that actual use exerts a significant indirect effect on tax compliance intention through perceived reduced compliance costs ($\beta$ = 0.290, p-value < 0.01). This finding illustrates how efficiency gains at the individual level—such as reduced administrative burden and time savings—can extend to the organizational level by fostering stronger compliance intentions. Collectively, these results highlight that both perceived usefulness and perceived cost reduction serve as key cognitive channels through which system interaction translates into meaningful compliance-oriented behavior.

5.5. Robustness Checks

Robustness assessment has become a standard practice in regression-based research, where scholars evaluate the stability of baseline coefficients under alternative model specifications—typically by adding or removing explanatory variables [177]. However, this important analytical step remains rarely implemented in empirical studies employing the PLS-SEM approach [175]. For example, Vaithilingam et al. [178] found that among 1228 PLS-SEM studies reviewed, only about 15% explicitly conducted robustness analyses. Moreover, even when such tests are performed, prior research has highlighted uncertainty not only regarding the appropriate techniques but also the contextual relevance of their application. In this regard, researchers are encouraged to adopt a rigorous and transparent approach rather than a selective one when conducting and reporting robustness checks, as failure to do so may undermine the credibility of their findings. In line with recent calls for enhanced robustness testing in PLS-SEM research [179], this study follows the recommendations of Sarstedt et al. [180] by examining three key sources of potential model misspecification—nonlinearity, endogeneity, and unobserved heterogeneity.

5.5.1. Nonlinear Effect

In path analysis, researchers often assume linear relationships among constructs. While this assumption generally reflects real-world phenomena, it does not always hold true [181]. To identify potential nonlinear effects, predictors can be transformed into quadratic terms to construct polynomial models [182]. A significant positive (or negative) quadratic term indicates that the predictor’s effect on the endogenous variable increases (or decreases) at higher predictor values, whereas a nonsignificant term suggests that the linear relationship remains robust. The quadratic effects function in SmartPLS facilitates this assessment efficiently and the results are detailed in

Table 10. Nonlinear Effect Assessment Results.

Relationship	$\mathit{\beta }$	Std. Deviation	t-Statistics	${\mathit{f}}^2$	Assessment
SysQ × SysQ → SF	−0.021	0.036	0.590	0.001	No Effect
SrvQ × SrvQ → SF	0.042	0.047	0.897	0.003	No Effect
InfQ × InfQ → SF	−0.043	0.040	1.077	0.004	No Effect
PEU × PEU → IU	0.002	0.039	0.061	0.000	No Effect
PEU × PEU → PU	−0.061	0.029	2.058 **	0.014	No Effect
PU × PU → IU	0.021	0.055	0.384	0.001	No Effect
SF × SF → IU	−0.024	0.063	0.378	0.001	No Effect
SF × SF → AU	0.022	0.065	0.336	0.001	No Effect
IU × IU → AU	−0.002	0.054	0.039	0.000	No Effect
AU × AU → RCC	−0.115	0.028	4.067 **	0.070	Small
RCC × RCC → TCI	−0.093	0.030	3.127 **	0.055	Small

Note: ** represents statistical significance at the 1% level.

.

Our analysis revealed that although most nonlinear effects were statistically insignificant, three quadratic relationships emerged as significant: between perceived ease of use and perceived usefulness ($\beta$ = −0.061, p-value < 0.01), between actual use and perceived reduced compliance costs ($\beta$ = −0.115, p-value < 0.01), and between perceived reduced compliance costs and tax compliance intention ($\beta$ = −0.093, p-value < 0.01). The effect sizes associated with these nonlinearities were small for actual use ($f^2=$ 0.070) and perceived reduced compliance costs ($f^2=$ 0.055), and negligible for perceived ease of use ($f^2=$ 0.014). Accordingly, we conclude that aside from these three relationships, the structural model remains robust under the assumption of linearity. These findings suggest that the associations between perceived ease of use and perceived usefulness, actual use and perceived reduced compliance costs, and perceived reduced compliance costs and tax compliance intention are not entirely linear. Instead, they display a more intricate behavioral pattern following an inverted U-shaped curve. Specifically, the initial impact of CTAS use on perceived reductions in compliance costs appears to rise up to a certain point, after which it gradually diminishes as the intensity and frequency of system use increase. This ‘rise-then-fall’ pattern similarly characterizes the other nonlinear relationships identified in the model.

5.5.2. Endogeneity

Endogeneity arises when a predictor correlates with the model’s error term [183], often due to the omission of relevant explanatory variables [184]. Such omitted variables become part of the error term and may correlate with the included predictors, thereby violating the exogeneity assumption and producing biased and inconsistent estimates [185]. Although endogeneity has long been recognized as a major concern in regression-based research [186], its discussion within the PLS-SEM literature remains limited [187]. Some scholars have mistakenly argued that PLS-SEM is less prone to endogeneity issues [188,189], even though the method relies on regression-based parameter estimation, where endogeneity remains a relevant concern [180]. To address this issue, both instrumental variable (IV) and non-IV approaches can be employed. However, the IV method often faces challenges due to the difficulty of identifying valid instruments [186]. Consequently, we adopt non-IV techniques, such as the Gaussian copula method [190], which accounts for endogeneity by explicitly modeling the correlation between endogenous predictors and the error term. The results derived from this method are summarized in

Table 11. Endogeneity Detection.

Relationship	$\mathit{\beta }$	Std. Deviation	t-Statistics	Detection
SysQ → SF (Endogenous Explanatories: CSysQ)	−0.141	0.187	0.755	No Endogeneity
SrvQ → SF (Endogenous Explanatories: CSrvQ)	0.090	0.100	0.898	No Endogeneity
InfQ → SF (Endogenous Explanatories: CInfQ)	−0.060	0.084	0.715	No Endogeneity
PEU → IU (Endogenous Explanatories: CPEU)	−0.090	0.069	1.307	No Endogeneity
PEU → PU (Endogenous Explanatories: CPEU)	−0.038	0.155	0.249	No Endogeneity
PU → IU (Endogenous Explanatories: CPU)	−0.091	0.068	1.333	No Endogeneity
SF → IU (Endogenous Explanatories: CSF)	−0.052	0.079	0.659	No Endogeneity
SF → AU (Endogenous Explanatories: CSF)	−0.058	0.058	1.009	No Endogeneity
IU → AU (Endogenous Explanatories: CIU)	−0.255	0.119	2.134 **	Endogeneity Exist
AU → RCC (Endogenous Explanatories: CAU)	−0.504	0.115	4.397 **	Endogeneity Exist
RCC → TCI (Endogenous Explanatories: CRCC)	−0.491	0.085	5.815 **	Endogeneity Exist

Note: ^C indicates the copula term in the model. ** represents statistical significance at the 1% level.

.

The results reveal that only three Gaussian copula terms—specifically those associated with intention to use (^CIU), actual use (^CAU), and perceived reduced compliance costs (^CRCC)—were statistically significant at the 1% level (p-value < 0.01). This finding indicates potential endogeneity concerns related to these constructs, likely stemming from unobserved factors that simultaneously shape users’ attitudes, behaviors, and efficiency perceptions toward CTAS. In contrast, the copula coefficients for system quality, service quality, information quality, perceived ease of use, perceived usefulness, and satisfaction (^CSysQ, ^CSrvQ, ^CInfQ, ^CPEU, ^CPU, and ^CSF) were all insignificant (p-value > 0.05), confirming that these variables can be treated as exogenous within the structural model. Consequently, apart from the pathways involving attitudinal, behavioral, and efficiency-related constructs, the estimated relationships appear robust and free from endogeneity bias. The presence of endogeneity in these three constructs is conceptually reasonable, as attitudes and behaviors often evolve through reciprocal causation, where prior system experiences reinforce perceptions of efficiency and shape subsequent behavioral intentions. Nonetheless, the existence of endogeneity along paths where these constructs act as predictors suggests that the corresponding relationships should be interpreted as indicative associations rather than definitive causal effects.

However, since the primary objective of the present study lies in assessing the predictive strength of the proposed variables rather than providing causal explanation of the model, endogeneity does not pose a major concern, as emphasized by Hair et al. [156]. Moreover, after incorporating the Gaussian copula terms for intention to use (^CIU), actual use (^CAU), and perceived reduced compliance costs (^CRCC), the corresponding path coefficients retained both the direction and statistical significance observed in the baseline estimation—indicating that the structural relationships remained stable following endogeneity adjustment. Nonetheless, the magnitude of these coefficients increased substantially—from 0.434 to 0.719 for intention to use, from 0.444 to 0.886 for actual use, and from 0.653 to 1.026 for perceived reduced compliance costs—suggesting that the original estimates were downward biased due to endogeneity (see

Table 12. Endogeneity-Adjusted Path Coefficients.

Relationship	Model	$\mathit{\beta }$	p-Value
IU → AU	Baseline results (see Table 7)	0.434	0.000
	After controlling for endogeneity	0.719	0.000
AU → RCC	Baseline results (see Table 7)	0.444	0.000
	After controlling for endogeneity	0.886	0.000
RCC → TCI	Baseline results (see Table 7)	0.653	0.000
	After controlling for endogeneity	1.026	0.000

).

5.5.3. Unobserved Heterogeneity

In the context of PLS-SEM, assuming that the population under investigation is homogeneous—an assumption that rarely holds true in business research—can result in misleading interpretations when data with underlying diversity are analyzed as a single group [191]. To account for heterogeneity, researchers may divide the sample into subgroups based on identifiable characteristics and estimate separate structural models for each, thus capturing observed heterogeneity. However, when the sources of variation are not known beforehand, it becomes necessary to explore unobserved heterogeneity within the data. If diagnostic evaluations suggest that unobserved heterogeneity does not meaningfully affect the model’s estimates, the dataset can be analyzed as a whole, allowing generalization of the findings. In contrast, if evidence of unobserved heterogeneity is found, appropriate methodological adjustments are required to account for it.

To detect and manage unobserved heterogeneity, PLS-SEM researchers can employ various latent class techniques, among which finite mixture PLS (FIMIX-PLS) is the most widely used [192]. The integration of this technique into major PLS-SEM software such as SmartPLS 4.1.1.2, coupled with extensive methodological guidance [193,194], has made its application accessible to researchers for years. According to Sarstedt et al. [191], FIMIX-PLS reliably identifies unobserved heterogeneity and determines the optimal number of latent segments based on several information criteria, including the modified Akaike information criterion with factor 3 (AIC3) [195] and the consistent AIC (CAIC) [196]. Additional considerations include segment ambiguity, which is reflected in high values of the entropy statistic (EN) [197]. Sarstedt et al. [192] further noted that AIC4 and the Bayesian information criterion (BIC) typically perform well in segment selection. Among these criteria, the minimum description length with factor 5 (MDL5) is regarded as the most robust, as it imposes a stronger penalty for model complexity [198]. If the indices consistently point to a one-segment solution or yield inconsistent segmentation results, researchers may reasonably conclude that unobserved heterogeneity does not exert a critical influence on the model [199].

Following the procedure outlined by Matthews et al. [194], the FIMIX-PLS analysis was initiated by assuming a one-segment solution, using the default settings for the stopping criterion (10^–7 = 1.0 × 10^–7), the maximum number of iterations (5000), and the number of repetitions (10). To determine the maximum number of segments to extract, we first calculated the minimum required sample size for estimating each segment, as recommended by Sarstedt et al. [199]. The results of a post hoc power analysis, assuming an effect size of 0.15 and a statistical power level of 80%, indicated a minimum sample size requirement of 85 per segment. This allowed for the extraction of up to four segments from our dataset. Accordingly, we re-ran the FIMIX-PLS procedure for two-, three-, and four-segment solutions using the same parameter settings as in the initial analysis. The results of these estimations are presented in

Table 13. Information Criteria for Segmentation Solutions.

	Number of Segments
Criteria	1	2	3	4
AIC3	5094.229	4664.935	4415.253	4183.514
AIC4	5111.229	4699.935	4468.253	4254.514
BIC	5143.007	4765.360	4567.325	4387.234
CAIC	5160.007	4800.360	4620.325	4458.234
MDL5	5542.119	5587.062	5811.616	6054.114
EN	N/A	0.947	0.759	0.778

Note: N/A = not available. Values in italics denote the smallest (i.e., best) outcome per segment retention criterion, as lower values of information criteria indicate a better model fit while balancing parsimony and explanatory power.

.

As shown in Table 13, most information criteria—namely AIC3, AIC4, BIC, and CAIC—consistently decrease as the number of segments increases from one to four, with the lowest values observed for the four-segment solution. This pattern suggests that models with a higher number of segments better capture data heterogeneity. Conversely, the MDL5 criterion reaches its minimum at the single-segment solution (MDL5 = 5542.119), implying that a one-segment model provides a sufficiently parsimonious representation of the data. Evidence from the relative segment sizes (

Table 14. Relative Segment Sizes (Observations = 354).

Number of Segments	Segment 1	Segment 2	Segment 3	Segment 4
1	1.000
2	0.817	0.183
3	0.491	0.368	0.140
4	0.476	0.277	0.138	0.110

Note: The table displays the relative sizes of the segments in descending order for each solution (i.e., per row). Due to the issue of label switching, a particular segment may be assigned different labels across alternative solutions [200].

) also aligns with the selection of the one-segment solution. Meanwhile, the entropy-based indices for the two-segment solution (EN = 0.947) indicate clear class separation, though not necessarily stronger model fit compared to the four-segment alternative. Jointly, the analyses do not unambiguously point to a specific segmentation solution, because (1) most information criteria (AIC variants, BIC, and CAIC) indicate a gradual improvement with additional segments, whereas (2) MDL5 favors the one-segment model. We therefore assume that unobserved heterogeneity is not at a critical level, which supports the validity of the overall-sample analysis.

6. Discussion

6.1. Theoretical Implications

Our results substantively extend the TAM and the D&M model by demonstrating how their constructs operate synergistically within the specific and policy-relevant context of digital tax administration. Consistent with the D&M framework, system quality, service quality, and information quality significantly and positively predict user satisfaction. Regarding system quality, this finding reinforces prior evidence suggesting that the absence of technical difficulties when navigating a website or delays in accessing web pages serves as a key antecedent of satisfaction in any e-government platform [201,202].

In terms of service quality, the results align with Saptono et al. [26], who emphasized that an electronic tax system that is responsive to service requests and provides personalized assistance to taxpayers enhances overall satisfaction. Similarly, with respect to information quality, our results corroborate Chen et al. [203], who found that when information provided by an IS is complete and accurate, users experience greater satisfaction because such information enables them to comply with tax obligations correctly and efficiently. Moreover, satisfaction derived from these quality dimensions exerts a strong positive effect on the intention to use the system, consistent with earlier empirical evidence demonstrating that satisfied users are more likely to continue or expand their engagement with digital systems [204,205].

In parallel, the core mechanism of TAM—where perceived ease of use enhances perceived usefulness, which in turn drives the intention to use—is strongly validated in this study. This finding is theoretically intuitive: a system that is easy to navigate reduces users’ cognitive load and learning costs, thereby reinforcing perceptions of its functional utility for accomplishing tasks more effectively [19,55]. Once users recognize these functional advantages, they are more likely to develop favorable behavioral intentions toward continued system use. This mechanism is consistent with extensive empirical research in both e-government [206,207] and e-tax contexts [208,209], as well as meta-analytic evidence by Dwivedi et al. [210], all of which confirm that perceived usefulness is the most proximal and influential determinant of technology adoption intention.

Furthermore, the analysis of indirect effects reveals a distinct pattern of asymmetric mediation across the two frameworks. Within the D&M model, intention to use functions as a pivotal mediator that translates evaluative constructs (system, service, and information quality, along with satisfaction) into behavioral outcomes such as actual use. The indirect pathways from quality constructs through satisfaction alone to actual use were not statistically significant; only when intention to use was introduced as a sequential mediator did the indirect effect become significant. In contrast, within TAM, perceived usefulness plays an analogous mediating role. Perceived ease of use cannot independently foster usage intention unless it is first internalized as perceived usefulness—a cognitive belief about the system’s utility. Once this belief is established, the pathway from ease of use to actual use becomes significant. The joint implication is theoretically meaningful: cognitive beliefs about utility and the behavioral intentions they generate form the crucial bridge between evaluative judgments and enacted usage routines [55,211]. This finding clarifies that, in post-adoption contexts, satisfaction and evaluative assessments alone are insufficient to produce observable behavioral engagement unless these judgments are transformed into motivational and cognitive mechanisms that sustain actual system use [212,213].

Beyond behavioral adoption, our findings suggest that higher levels of actual use—reflected in both the frequency and intensity of CTAS engagement—are significantly associated with stronger perceptions of reduced compliance costs. These perceptions capture the tangible individual-level benefits experienced by tax professionals. For tax consultants, CTAS minimizes the administrative burden of assisting multiple clients by streamlining return filing, validation, and reporting processes. For tax accountants, the system reduces time pressure and procedural complexity in corporate tax management, allowing them to allocate more effort toward strategic compliance planning. These perceived efficiency gains appear to contribute to professional satisfaction and may reinforce individual commitment to accurate and timely tax practices [135,214]. At a higher level, such individual-level benefits aggregate into organizational advantages: clients and employers alike experience lower risks of reporting errors, noncompliance, and penalties. In this way, tax professionals’ intention to comply—strengthened through their digital engagement—appears to function as a behavioral bridge between technological adoption and organizational compliance integrity [215,216]. Consequently, the study extends existing theory by clarifying how digital tax systems generate multi-level benefits, enhancing both individual efficiency and collective compliance outcomes.

In addition, the nonlinearity assessment provides further theoretical refinement by revealing that several key structural relationships in the model are not strictly linear. Specifically, the significant quadratic terms indicate that initial increases in perceived ease of use, actual use, and perceived reductions in compliance costs are associated with progressively stronger downstream relationships with perceived usefulness, perceived compliance cost reductions, and tax compliance intentions, respectively—up to an optimal point. Beyond this point, however, marginal returns diminish and may eventually reverse, producing an inverted U-shaped pattern. In the context of perceived ease of use, this finding aligns with Csikszentmihalyi’s [217] flow theory, which posits that once an optimal threshold of ease is reached, further simplification may induce boredom, leading users to lose interest and engage with the website only sporadically, thereby reducing its incremental contribution to perceived usefulness. Empirical studies in human–computer interaction have demonstrated similar dynamics; for instance, Sánchez-Franco and Roldán [218] and Sharif and Naghavi [219] integrated the flow principles into TAM and showed that effective system design requires not only ease but also an appropriate level of challenge and stimulation to sustain user absorption and perceived utility.

A comparable inverted U-shaped mechanism also emerges in the relationships involving actual use and perceived reductions in compliance costs, which can be interpreted through the lens of work knowledge and task proficiency. Foundational studies in industrial-organizational psychology [220,221,222] and the tenure–performance meta-analysis by Sturman [223] demonstrate that performance gains from experience are disproportionately large at early stages but taper over time as fewer new skills remain to be mastered. Applied to CTAS, intensive early use provides substantial efficiency benefits—faster navigation, fewer clerical errors, and smoother filing procedures—which strongly reinforce perceptions of reduced compliance costs. Yet, as tax professionals become highly proficient and system interaction becomes routine, additional usage yields smaller gains. In some cases, overreliance on repetitive or highly routinized digital interactions may even foster complacency, overconfidence, or reduced attentiveness [224,225], thereby weakening perceptions of cost savings.

With respect to the relationship between perceived reductions in compliance costs and tax compliance intentions, a similar inverted U-shaped dynamic indicates that beyond a certain point, further reductions in perceived compliance burden may inadvertently undermine compliance motivation. When compliance becomes excessively effortless, users may shift from careful verification to superficial processing, relying too heavily on automated features such as prepopulated returns and system-generated validations. Prior studies—including Fonseca and Grimshaw [226], van Dijk et al. [227], Doxey et al. [228], and Fochmann et al. [229]—show that such overautomation can reduce vigilance, increase omission errors, and weaken accuracy-oriented behaviors, consistent with omission bias theory [230,231,232]. This pattern is also consistent with motivation crowding theory [233], which posits that excessive external facilitation can crowd out intrinsic motivations for due diligence. In the tax context, this means that when systems become too frictionless, they may inadvertently erode users’ internalized commitment to complete, accurate, and carefully reviewed reporting. Collectively, these nonlinear patterns highlight the importance of balancing efficiency with mechanisms that preserve professional vigilance and intrinsic compliance motivations.

Beyond those theoretical contributions, this study advances the methodological rigour of IS and digital tax administration research by explicitly addressing two robustness concerns that are often overlooked in empirical studies: endogeneity and unobserved heterogeneity [187,199]. Our endogeneity assessment revealed a downward bias in the baseline estimates within the intention to use–actual use–perceived compliance cost subsystem, demonstrating that causal inferences in post-adoption models can be systematically distorted when reciprocal relationships or omitted variables are ignored. Likewise, our heterogeneity diagnostics indicated that the structural relationships tested were largely stable across unobserved subpopulations, thereby strengthening the generalizability of the findings while simultaneously illustrating the value of formally testing for latent segmentation rather than assuming population homogeneity. These robustness checks contribute to the literature by showing that digital tax administration research—much like broader IS scholarship—benefits from moving beyond traditional linear PLS-SEM approaches toward models that explicitly account for endogeneity and latent heterogeneity. In doing so, this study sets a methodological precedent and encourages future work in e-tax and e-government systems to incorporate similar diagnostic procedures to ensure the credibility, accuracy, and replicability of empirical conclusions.

6.2. Practical Implications

The findings of this study offer several practical implications for policymakers, businesses, and tax professionals seeking to ensure the successful implementation of CTAS as a key component of Indonesia’s digital tax administration reforms. First, because system, service, and information quality jointly shape user satisfaction—and satisfaction in turn reinforces behavioral intention—policymakers must prioritise system reliability, responsiveness, and informational accuracy in the ongoing development of CTAS. Reducing system downtime, improving response speed (latency) during peak filing periods (typically in March), and enhancing the clarity and completeness of information—particularly for complex transactions—are essential for sustaining users’ evaluative trust in the platform.

These imperatives manifest differently across professional user groups. For tax consultants, who manage large and diverse client portfolios with varied compliance timelines, even minor system disruptions can cascade into significant operational delays, making high system stability and consistent performance indispensable for maintaining service delivery to their clients [234]. Tax accountants, although equally dependent on system reliability, face a distinct set of demands: their work hinges on the accuracy of information [235], smooth data reconciliation processes, and the consistent functioning of technical features such as prepopulated forms, automated validation routines, and integrated data flows. Minor informational inconsistencies or validation irregularities can disrupt monthly and annual closing cycles, elevate the risk of reporting errors, and impose additional coordination burdens across internal organizational units.

Second, given that intention to use functions as the pivotal mediator between satisfaction and actual use within the D&M model, implementation strategies must actively convert positive evaluations into sustained behavioral commitment. Goal-oriented nudges (such as deadline reminders that highlight task-specific benefits), guided workflows that break filing procedures into salient milestones, and commitment devices such as integrated filing schedulers can strengthen intention formation. On the institutional side, tax authorities may support this process by establishing a specialised CTAS management unit—staffed with highly trained personnel—to monitor system performance, respond rapidly to reported issues, and maintain ongoing communication with professional users [26]. From the organizational perspective, both firms and tax advisory practices can reinforce intention formation by embedding CTAS usage expectations into job descriptions, standard operating procedures, and performance evaluation systems.

Third, in line with the TAM mechanism in which perceived usefulness is the principal conduit through which perceived ease of use shapes behavioral intention, system refinements should prioritise task-relevant utility over cosmetic usability enhancements. Features that demonstrably improve accuracy, reduce processing time, and support productivity—such as reliable calculation modules, contextual guidance for complex filings, and actionable compliance dashboards—are more likely to strengthen behavioral intention than interface adjustments alone. Accordingly, communication campaigns and training programs should emphasise instrumental benefits such as time savings, error reduction, and enhanced audit readiness.

Fourth, the nonlinear patterns uncovered in the analysis—particularly the inverted U-shaped effects—carry significant practical implications for system design and policy experimentation. For perceived ease of use, simplification beyond an optimal threshold may diminish perceived usefulness by reducing user engagement and attentiveness. Policymakers should therefore avoid excessive automation that eliminates meaningful user interaction or oversight. Similarly, the nonlinearities involving actual use and perceived reductions in compliance costs caution against overly routinized workflows. To preserve vigilance among experienced users, CTAS should incorporate periodic prompts requiring manual review, randomized checks, or reflective confirmations—interventions shown to reduce complacency and omission errors [224,227]. Such design safeguards are especially important for tax consultants and tax accountants whose professional responsibilities require high levels of verification, even in digital environments.

Finally, the endogeneity detected along the sequential pathway linking intention to use, actual use, perceived reductions in compliance costs, and tax compliance intentions has direct practical relevance. Because these constructs are shaped by simultaneous and mutually reinforcing processes—rather than by strictly unidirectional causal pathways—interventions targeting any point along this chain can generate cascading effects throughout the compliance process. Early-stage engagement measures such as structured onboarding, first-filing assistance, or targeted support for complex transactions are therefore likely to produce substantial downstream benefits by enhancing actual use, strengthening perceived efficiency gains, and ultimately reinforcing compliance intentions. The resulting increase in perceived benefits, in turn, reinforces users’ intention to continue using CTAS, thereby sustaining higher levels of actual system use over time. Such feedback effects would remain hidden without formal endogeneity diagnostics, highlighting the strategic importance of disproportionately investing in early-cycle interactions where reciprocal gains are strongest.

Meanwhile, the limited unobserved heterogeneity observed across users suggests that these interventions can be broadly implemented for both tax accountants and tax consultants, although future refinements may tailor support to more specialised subgroups (e.g., industry-specialised accountants or sole-practitioner consultants). For this reason, periodic diagnostic assessments of endogeneity and latent segmentation should be institutionalised to ensure that CTAS enhancements continue to deliver system-wide benefits across diverse professional profiles.

7. Conclusions

This study evaluates the early-stage effectiveness of Indonesia’s newly implemented CTAS by integrating two foundational IS frameworks: the D&M model and the TAM. Using survey data from 354 tax professionals—tax consultants and tax accountants who represent the system’s most intensive and technically proficient users—the study tests eleven hypotheses, capturing perceptions, usage behavior, and compliance-related outcomes.

The findings underscore the central role of satisfaction and perceived usefulness in shaping behavioral intention. System quality, service quality, and information quality contribute positively—though more modestly—to satisfaction, while perceived ease of use strongly enhances both perceived usefulness and intention to use. Actual system use, in turn, strengthens perceptions of reduced compliance costs, which emerge as a key driver of tax compliance intention. Robustness analyses further refine these insights by revealing meaningful nonlinearities and identifying endogeneity along the sequential pathway linking intention to use, actual use, perceived compliance cost reductions, and compliance intention. These results demonstrate that post-adoption perceptions and behaviors are mutually reinforcing rather than strictly unidirectional, contributing important empirical evidence to IS and digital tax administration research.

Several limitations should be acknowledged. The cross-sectional design restricts causal inference and limits insight into dynamic learning and adaptation processes. Reliance on self-reported measures may introduce perceptual bias and does not capture objective usage or compliance outcomes. In addition, the use of convenience sampling may constrain the external validity of the findings, as the sample consists exclusively of tax professionals who represent highly experienced and intensive users of the system. Consequently, the results should be interpreted with caution when generalizing to the broader population of taxpayers in Indonesia, whose technological familiarity, compliance motivations, and interactions with the tax administration system may differ.

Furthermore, although endogeneity was explicitly tested using the Gaussian copula approach, the presence of endogeneity signals along parts of the behavioral pathway suggests that the estimated relationships should be interpreted with caution. Unobserved factors or reciprocal influences among constructs may still affect the magnitude of the estimated effects.

In addition, although the dataset includes several demographic characteristics of respondents, this study does not explicitly examine potential differences in technology acceptance across demographic groups such as gender, professional experience, or educational background. While unobserved heterogeneity was assessed using the FIMIX-PLS approach, which indicates that the dataset can be adequately represented by a single population structure, future research could further investigate observed demographic heterogeneity to better understand how different professional profiles perceive and adopt digital tax administration systems.

These limitations point to several directions for future research. Longitudinal designs could capture evolving perceptions, learning effects, and reciprocal behavioral mechanisms as users gain experience with CTAS. Future studies may also benefit from incorporating objective indicators—such as administrative records or system-log data—to complement perceptual measures and provide stronger evidence of actual system use and compliance outcomes. Expanding the analysis beyond tax professionals to include a broader population of taxpayers would further enhance the external validity of the findings. Moreover, additional research could explore demographic or professional heterogeneity in technology acceptance, as well as examine CTAS performance across industries with varying compliance complexity. As CTAS continues to mature, such research will be essential for building a cumulative and policy-relevant evidence base on digital tax administration in emerging economies.